Mimicking titration experiments with MD simulations: A protocol for the investigation of pH-dependent effects on proteins

www.nature.com/scientificreports OPEN received: 24 September 2015 accepted: 15 February 2016 Published: 03 March 2016 Mimicking titration experiments with MD simulations: A protocol for the investigation of pH-dependent effects on proteins Eileen Socher & Heinrich Sticht Protein structure and function are highly dependent on the environmental pH. However, the temporal or spatial resolution of experimental approaches hampers direct observation of pH-induced conformational changes at the atomic level. Molecular dynamics (MD) simulation strategies (e.g. constant pH MD) have been developed to bridge this gap. However, one frequent problem is the sampling of unrealistic conformations, which may also lead to poor pKa predictions. To address this problem, we have developed and benchmarked the pH-titration MD (pHtMD) approach, which is inspired by wet-lab titration experiments. We give several examples how the pHtMD protocol can be applied for pKa calculation including peptide systems, Staphylococcus nuclease (SNase), and the chaperone HdeA. For HdeA, pHtMD is also capable of monitoring pH-dependent dimer dissociation in accordance with experiments. We conclude that pHtMD represents a versatile tool for pKa value calculation and simulation of pH-dependent effects in proteins. Solution pH can have a drastic effect on protein structure and function, which has been exploited by nature to trigger a large variety of physiological processes. For example, some bacteria are able to survive the acidic conditions in the stomach of their host by using acid-activated chaperones which protect substrate proteins upon binding1. In viruses, some of the fusion proteins that mediate cell entry have been described to act pH-dependently2,3. Other proteins in vertebrates undergo pH changes during their maturation on the way through the endoplasmic reticulum and the Golgi apparatus4. In plants, the simultaneous closure of water channels has been observed as a response to changing pH values during flooding5. On a molecular level, changes in the pH value affect the protonation state of several types of amino acids, including aspartate, glutamate, histidine, lysine, cysteine, and tyrosine. The addition or removal of a proton always changes the charge of the respective amino acid side chain, thereby affecting the charge distribution within the protein, which may lead to conformational changes. For instance, these structural alterations can trigger changes in protein activity, ligand binding properties, or the oligomerization state. However, due to the temporal or spatial resolution of experimental approaches, it is extremely difficult to observe pH-induced conformational changes in proteins directly at the atomic level. Also the generation of structural data at different pH values, for instance with X-ray crystallography or NMR spectroscopy, underlies different restrictions and is technically very demanding. To mention only a few general limitations, proteins mostly do not crystallize at very different pH values and NMR spectroscopy is limited to small proteins. At this point, molecular dynamics (MD) simulations, which start from experimentally determined structures, can help investigate the effect of pH changes on an atomic level and on picosecond to microsecond time scales. One hallmark of conventional MD simulations is the fact that an initially assigned protonation state cannot be changed during the simulation. This “constant protonation” approach results in some drawbacks for studying pH-dependent effects6: (1) Assigning the right protonation states for the titratable groups in the protein requires knowledge of their pKa values, (2) if any of these pKa values are near the solvent pH there may be no single protonation state that adequately represents the ensemble of protonation states appropriate at that pH, and (3) the invariable protonation states decouple the dynamic dependence of pKa and protonation state on conformation. Division of Bioinformatics, Institute of Biochemistry, Friedrich-Alexander-University Erlangen-Nürnberg (FAU), Fahrstraße 17, 91054 Erlangen, Germany. Correspondence and requests for materials should be addressed to H.S. (email: ) Scientific Reports | 6:22523 | DOI: 10.1038/srep22523 1 www.nature.com/scientificreports/ Figure 1. Workflow of pHtMD simulations. The pHtMD simulation starts with a model compound or an experimentally determined structure. At first, a 1 ns long CpHMD simulation is performed (blue line). The final coordinates and velocities are transferred (dashed orange lines) to serve as a starting point for the next 1 ns long CpHMD simulation (blue lines), which has now a slightly lowered pH compared to the previous 1 ns. These steps are repeated until the final pH value is reached. This example shows a systematic lowering of the pH; a systematic increase of the pH can be accomplished in an analogous fashion. The data obtained from the pHtMD can be analyzed with respect to different aspects, for instance pKa values, conformational features or net charges of proteins. To avoid these problems, the constant pH molecular dynamics (CpHMD) approach was developed6,7. One widespread implementation, for example in the AMBER software suite, performs Monte Carlo sampling of the Boltzmann distribution of protonation states interspersed in the molecular dynamics simulation8. Thereby, the solution pH is set as an external variable determining the distribution of the different protonation states, which are modeled by different charge sets8. CpHMD has become a popular method to study the pH-dependence of protein9 and peptide10 structures or to calculate the pKa values of titratable residues6,11. However, a comparison between calculated and experimentally determined pKa values frequently revealed significant differences indicating that unrealistic protein conformations are sampled11,12. Recent approaches to reduce this problem are constant pH replica exchange molecular dynamics (pH-REMD) simulations13,14 and the explicit consideration of the solvent12,15. As an alternative approach, we have devised a modified procedure, which is inspired by wet-lab titration experiments. This pH-titration MD (pHtMD) relies on the overall concept of CpHMD, but performs a consecutive series of MD simulations with small pH changes, which allows a smooth adaption of the structure to the solvent pH (Fig. 1). The rationale for suggesting this titration concept was the following: Conventional CpHMD usually runs a set of simulations at different pH values that are fixed at the beginning of each simulation and may differ significantly from the pH at which the structure was determined (e.g. pH 3 simulation using a pH 8 structure as a template). CpHMD thus requires a rapid adaptation of the structure to different pH values, which may cause the sampling of unrealistic conformations thereby producing inaccurate pKa values. To address this problem, we have developed and benchmarked the pHtMD approach (...truncated)